in reply to Module installation can be harmful

Upgrades have consequences. I can't speak for the 5.6.0 -> 5.6.1 Test::Harness problem but I do recall that you cannot just upgrade to 5.8.0 and expect compiled modules to work. Some binary APIs were changed or something - it's in the ChangeList or somewhere prominent that I know I read about it. All that means is that if you upgrade to 5.8 you can expect to have to somehow conspire for either all the compiled stuff to get recompiled or just install them over again.

In addition to that - keep in mind that any fool can upload new things to CPAN. People can't overwrite each other's work but there's nothing to stop someone from sending /new/ things up. This means there's some inherant trust issues regarding modules from CPAN. You have to trust that the author is both cluefull and benign. I do place a lot of trust in the various authors who publish to CPAN. The only alternative is to either audit all the code I download or just not download it. Both are fairly poor alternatives.

There is one mitigating factor: CPAN testers. There's already a great bunch of people out there downloading and trying new versions on their various machines. There's at least a chance that any problems will be at least noticed if not caught by these folks. If you follow the URL I just posted then you'll also notice an archived e-mail list of testing results. Those results are gathered back into CPAN proper and are visible when viewing the CPAN archive from search.cpan.org.

This leaves with some options - you can either approach CPAN with the appropriate wariness and use the tools provided (CPAN tester's results, is it a common module, ChangeLog info on incompatible changes) or you can continue to operate in a fantasy land where every new change and tweak is backwards compatible and is bug free.

__SIG__
printf "You are here %08x\n", unpack "L!", unpack "P4", pack "L!", B::
+svref_2object(sub{})->OUTSIDE

[download]

Replies are listed 'Best First'.
Re: Re: Module installation can be harmful
by toma (Vicar) on Oct 10, 2002 at 06:16 UTC
    This wasn't just any module, it is in the perl core.

    I didn't expect perl 5.8 to have binary compatibility with perl 5.6.X.

    One of the challenges with testing is trying things on the many different versions of systems. Imagine CPAN test results as a table, where the rows of the table are the modules, and the columns of the table are all the possible configurations of perl, I think that only a very small percentage of the table is filled in. It's a really great project and I hope that the table continues to become more filled-in.

    Yet I'm not sure that the cpan testers would have found this particular problem with an automated CPAN testing process. The module passed its tests. It just broke perl when I installed it. In a response to chromatic's node below, I have included more details about what happened.

    I don't expect every module to be backwards compatible. But if a pure perl module is in the core, and passes its tests, I don't expect it to break perl itself when it is installed. For the most part, my expectation has been met. I think that there is a big challenge to make a transition such as 5.6.0/1 to 5.6.8 work smoothly, and perhaps some modules have trouble with this.

    It should work perfectly the first time! - toma

[reply]

In Section Meditations